Deep learning approach to computing spans

A method of training a natural language processor, the method comprising:

• importing a highlighted training text including a first plurality of training nodes;
• importing a non-highlighted training text including a second plurality of training nodes;
• converting the highlighted training text into highlighted training conversion tables;
• converting the non-highlighted training text non-highlighted training conversion tables;
• training an enhanced-one-hot encoder using the highlighted and non-highlighted training conversion tables;
• enhanced-one-hot encoding the highlighted training conversion tables to generate highlighted training vectors;
• enhanced-one-hot encoding the non-highlighted conversion tables to generate non-highlighted training vectors,
• wherein enhanced-one-hot encoding comprises:
• generating a parse tree having a plurality of nodes;
• and
• generating a vector table that includes a first row and a second row, wherein:
• the first row represents a first traversal through the parse tree from a trigger node to a first target node, wherein the trigger node includes a first attribute and a second attribute, and wherein the first target node includes a third attribute and a fourth attribute;
• and
• the second row represents a second traversal through the parse tree from the trigger node to a second target node, wherein the second target node includes a fifth attribute and a sixth attribute;
• wherein the column headings of the vector table comprise a plurality of attributes including the first attribute, the second attribute, the third attribute, the fourth attribute, the fifth attribute, and the sixth attribute;
• wherein each position in the first row of the vector table includes a “1” for each column heading that is true for the first traversal;
• and
• wherein each position in the first row of the vector table includes a “0” for each column heading that is false for the first traversal;
• training a projection model using the highlighted and non-highlighted training vectors;
• processing the highlighted training vectors using the projection model to generate highlighted processed training vectors;
• and
• training a classifier model using the highlighted processed training vectors,
• wherein the classifier model determines whether a node is in a sought-after class.

A system to find nodes in a span, the system comprising:

• a plurality of highlighted parse trees representing labeled natural language text;
• a plurality of non-highlighted parse trees representing unlabeled natural language text;
• a new parse tree representing new natural language text;
• a natural language processing (NLP) learning machine configured to process the plurality of highlighted parse trees, the plurality of non-highlighted parse trees, and the new parse tree, wherein the NLP learning machine includes a computing processor;
• and
• a memory coupled to the computing processor, wherein the memory comprises instructions which, when executed by the computing processor, specifically configures the computing processor and causes the computing processor to:
• import a highlighted training text including a first plurality of training nodes;
• import a non-highlighted training text including a second plurality of training nodes;
• enhanced-one-hot encode the highlighted and non-highlighted training text, wherein enhanced-one-hot encoding comprises:
• generating a parse tree having a plurality of nodes;
• and
• generating a vector table that includes a first row and a second row, wherein:
• the first row represents a first traversal through the parse tree from a trigger node to a first target node, wherein the trigger node includes a first attribute and a second attribute, and wherein the first target node includes a third attribute and a fourth attribute;
• and
• the second row represents a second traversal through the parse tree from the trigger node to a second target node, wherein the second target node includes a fifth attribute and a sixth attribute;
• wherein the column headings of the vector table comprise a plurality of attributes including the first attribute, the second attribute, the third attribute, the fourth attribute, the fifth attribute, and the sixth attribute;
• wherein each position in the first row of the vector table includes a “1” for each column heading that is true for the first traversal;
• and
• wherein each position in the first row of the vector table includes a “0” for each column heading that is false for the first traversal;
• train a projection model using the highlighted and non-highlighted training text;
• processing the highlighted training text using the projection model;
• train a classifier model using the highlighted processed training text;
• import new text including a plurality of new nodes;
• enhanced-one-hot encode the new text;
• process the new text using the projection model;
• and
• determine, using the classifier model, whether one of the plurality of new nodes is in a sought-after class.

A system to find nodes in a span, the system comprising:

• a plurality of highlighted parse trees representing labeled natural language text;
• a plurality of non-highlighted parse trees representing unlabeled natural language text;
• a new parse tree representing new natural language text;
• a natural language processing (NLP) learning machine configured to process the plurality of highlighted parse trees, the plurality of non-highlighted parse trees, and the new parse tree, wherein the NLP learning machine includes a computing processor;
• and a memory coupled to the computing processor, wherein the memory comprises instructions which, when executed by the computing processor, specifically configures the computing processor and causes the computing processor to:
• convert the highlighted training text into highlighted training conversion tables;
• convert the non-highlighted training text non-highlighted training conversion tables;
• train an enhanced-one-hot encoder using the highlighted and non-highlighted training conversion tables;
• enhanced-one-hot encode the highlighted training conversion tables to generate highlighted training vectors,
• wherein enhanced-one-hot encoding comprises:
• generating the new parse tree having a plurality of nodes;
• and
• generating a vector table that includes a first row and a second row, wherein:
• the first row represents a first traversal through the new parse tree from a trigger node to a first target node, wherein the trigger node includes
• a first attribute and a second attribute, and wherein the first target node includes a third attribute and a fourth attribute;
• and
• the second row represents a second traversal through the new parse tree from the trigger node to a second target node, wherein the second target node includes a fifth attribute and a sixth attribute;
• wherein the column headings of the vector table comprise a plurality of attributes including the first attribute, the second attribute, the third attribute, the fourth attribute, the fifth attribute, and the sixth attribute;
• wherein each position in the first row of the vector table includes a “1” for each column heading that is true for the first traversal;
• and
• wherein each position in the first row of the vector table includes a “0” for each column heading that is false for the first traversal;
• enhanced-one-hot encode the non-highlighted conversion tables to generate non-highlighted training vectors,
• train a projection model using the highlighted and non-highlighted training vectors;
• process the highlighted training vectors using the projection model to generate highlighted processed training vectors;
• and
• train a classifier model using the highlighted processed training vectors,
• wherein the classifier model determines whether a node is in a sought-after class.

A computer program product comprising a computer readable storage medium having a computer readable program stored therein to find nodes in a span, wherein the computer readable program, when executed on a computing device, specifically configures the computing device, and causes the computing device to:

• import a highlighted training text including a first plurality of training nodes;
• import a non-highlighted training text including a second plurality of training nodes;
• enhanced-one-hot encode the highlighted and non-highlighted training text, wherein enhanced-one-hot encoding comprises:
• generating a parse tree having a plurality of nodes;
• and
• generating a vector table that includes a first row and a second row, wherein:
• the first row represents a first traversal through the parse tree from a trigger node to a first target node, wherein the trigger node includes a first attribute and a second attribute, and wherein the first target node includes a third attribute and a fourth attribute;
• and
• the second row represents a second traversal through the parse tree from the trigger node to a second target node, wherein the second target node includes a fifth attribute and a sixth attribute;
• wherein the column headings of the vector table comprise a plurality of attributes including the first attribute, the second attribute, the third attribute, the fourth attribute, the fifth attribute, and the sixth attribute;
• wherein each position in the first row of the vector table includes a “1” for each column heading that is true for the first traversal;
• and
• wherein each position in the first row of the vector table includes a “0” for each column heading that is false for the first traversal;
• train a projection model using the highlighted and non-highlighted training text;
• processing the highlighted training text using the projection model;
• train a classifier model using the highlighted processed training text;
• import new text including a plurality of new nodes;
• enhanced-one-hot encode the new text;
• process the new text using the projection model;
• and
• determine, using the classifier model, whether one of the plurality of new nodes is in a sought-after class.